Comparison of different fft-implementations

Specific Domains Signal and Image Processing
, ,
2

Works for me:

julia> x = rand(ComplexF64, 47499754);

julia> fft(x)
47499754-element Vector{ComplexF64}:
 2.3752510813742287e7 + 2.3752082562739518e7im
   -1857.327613402386 + 2656.595755623108im
   1.2335114129446083 - 1764.624248895477im
  -1396.5209606987896 + 1594.6473092040806im
   -994.9367333881451 - 693.3802670885018im
  -2928.6763801313805 - 1549.0585642899605im
    4545.894246165899 - 1879.5461946803975im
   1263.5570298623447 + 82.52881908200902im
  -1794.6733211109129 + 3696.6978993992598im
   -2795.508384720929 + 2427.6278018923563im
                      ⋮

If you use plan_fft with FFTW.MEASURE or FFTW.PATIENT, realize that it overwrites the data with zeros — this is a FAQ:

You should initialize your input array after creating the plan, unless you use FFTW_ESTIMATE : planning with FFTW_MEASURE or FFTW_PATIENT overwrites the input/output arrays, as described in the manual.

(Realize also that 47499754 is 2 \times 23749877 where 23749877 is prime. FFTW and recent versions of SciPy support large prime factors with O(n \log n) algorithms, but sizes with small prime factors will usually be at least 10\times more efficient.)

1 Like